海湾型半城市化地区空间形态演化模拟

Peri-Urban Landscape Dynamics Simulation by Integrating Simulated Annealing and Cellular Automata

元胞自动机(CA)是模拟城市土地利用演变过程的有效工具,转换规则和元胞邻域是元胞模型的核心。综合考虑元胞邻域的距离衰减效应,基于模拟退火算法(SA)挖掘最优的转换规则,文章构建了一种考虑邻域衰减的城市演化模型(SA-NDCA)。模型以负幂指数函数作为元胞邻域的衰减曲线表示元胞邻域的距离衰减效应;运用模拟退火优化算法计算城市CA模型模拟结果与样本点的累积差异,在目标解空间快速搜索以提取最优的转换规则;最后以厦门市半城市化地区为研究案例,模拟了研究区域1995―2010年期间的城市空间形态演化,通过混淆矩阵和Kappa系数评价了模型的模拟精度,1995―2010年期间的建设用地模拟精度为68.5%,总体精度达到86.2%,Kappa系数达到66.3,取得了较好的模拟效果。利用提出的SA-NDCA模型,成功模拟了研究区2010―2020年期间的城市空间形态演化,结果显示,所预测的演化情景与中国当前实施的新型城镇化战略十分契合。

Cellular Automata has proved to be a convenient and effective tool for simulating urban land use change, in which transition rules and neighborhoods configuration are the key issues. This paper presents a neighbor decay cellular automata model based on Simulated Annealing（SA-NDCA）, which uses simulated annealing（SA） to find transition rules and considers the decay effect of the cellular neighborhood. A negative power exponential function is used to compute the decay coefficient of cellular neighborhood in the model. With calculating the cumulative differences in simulating results and the sample data, the simulated annealing optimization automatically searches the optimal combination of parameters of transition rules. By using Xiamen City as a study case, the peri-urban landscape dynamics of the period 1995-2010 is simulated with the model. For further evaluating the model performance, this study computed model accuracy by confusion matrix and Kappa coefficient. During 1995-2010, the simulation accuracy for built-up land was 68.5%, and the overall accuracy was 86.2%, while Kappa coefficient reached 66.3. It was a relatively high level as compared with other researches. This study shows that the simulation results of SA-NDCA model can achieve a relatively high prediction accuracy of construction land, overall accuracy and Kappa coefficient. The neighbor decay cellular automata model based on simulated annealing optimization is very suitable for simulating the land-use changes in peri-urban area, as the model simulated well the characteristics of the intensive distribution of construction land development in the study area.